Datasets:
Add comprehensive dataset README
Browse files
README.md
CHANGED
|
@@ -38,20 +38,6 @@ pretty_name: FlareBench
|
|
| 38 |
|
| 39 |
**FlareBench** is a novel benchmark dataset for solar flare prediction that covers the entire 11-year solar activity cycle. This dataset was introduced in our ICCV 2025 paper "Deep Space Weather Model: Long-Range Solar Flare Prediction from Multi-Wavelength Images".
|
| 40 |
|
| 41 |
-
## 🚀 What's New
|
| 42 |
-
|
| 43 |
-
- **ICCV 2025 Accepted!** Our paper has been accepted to ICCV 2025
|
| 44 |
-
- **Complete Solar Cycle Coverage**: First benchmark covering a full 11-year solar activity cycle
|
| 45 |
-
- **Multi-Modal Data**: Combines solar images (AIA/HMI) and X-ray sensor data (XRS)
|
| 46 |
-
- **Time-Series Cross-Validation**: Designed for robust evaluation across diverse solar activity states
|
| 47 |
-
|
| 48 |
-
## 📊 Dataset Overview
|
| 49 |
-
|
| 50 |
-
FlareBench addresses the limitations of conventional solar flare prediction datasets that often:
|
| 51 |
-
- Cover only limited periods of solar activity
|
| 52 |
-
- Contain small, low-resolution sunspot patches
|
| 53 |
-
- Exhibit biases towards specific solar activity periods
|
| 54 |
-
|
| 55 |
### Why FlareBench?
|
| 56 |
|
| 57 |
Most conventional datasets for solar flare prediction do not cover diverse solar activity states. Consequently, models trained on such datasets can exhibit biases towards specific periods of solar activity. Furthermore, many datasets contain only small, low-resolution sunspot patches.
|
|
@@ -68,16 +54,16 @@ FlareBench/
|
|
| 68 |
├── solar_images/
|
| 69 |
│ ├── aia/ # Atmospheric Imaging Assembly (9 wavelengths)
|
| 70 |
│ │ ├── 2011/ # 📊 Data available (Apr-Dec 2011)
|
| 71 |
-
│ │ ├── 2012/ #
|
| 72 |
│ │ ├── ...
|
| 73 |
│ │ └── 2024/
|
| 74 |
│ └── hmi/ # Helioseismic and Magnetic Imager
|
| 75 |
│ ├── 2011/ # 📊 Data available (Jan-Dec 2011)
|
| 76 |
-
│ ├── 2012/ #
|
| 77 |
│ └── ...
|
| 78 |
└── xrs/ # X-Ray Sensor data from GOES satellites
|
| 79 |
├── 2011/ # 📊 Data available (full year)
|
| 80 |
-
├── 2012/ #
|
| 81 |
└── ...
|
| 82 |
```
|
| 83 |
|
|
@@ -123,15 +109,6 @@ We will provide access to the complete dataset for research purposes. Please inc
|
|
| 123 |
- Intended use of the dataset
|
| 124 |
- Preferred data transfer method
|
| 125 |
|
| 126 |
-
## 🔬 Evaluation Protocol
|
| 127 |
-
|
| 128 |
-
FlareBench uses **time-series cross-validation** to ensure unbiased evaluation:
|
| 129 |
-
|
| 130 |
-
- **3-fold cross-validation** covering different phases of the solar cycle
|
| 131 |
-
- **Chronological splits** maintaining temporal order
|
| 132 |
-
- **Diverse solar conditions** in each fold's test set
|
| 133 |
-
- **Standard metrics**: TSS, BSS, GMGS for comprehensive evaluation
|
| 134 |
-
|
| 135 |
## 📚 Citation
|
| 136 |
|
| 137 |
If you use FlareBench in your research, please cite our ICCV 2025 paper:
|
|
@@ -169,6 +146,3 @@ For questions, issues, or collaboration opportunities:
|
|
| 169 |
- **Institution**: Keio University
|
| 170 |
- **Project Page**: [https://keio-smilab25.github.io/DeepSWM](https://keio-smilab25.github.io/DeepSWM)
|
| 171 |
|
| 172 |
-
---
|
| 173 |
-
|
| 174 |
-
*Built with ❤️ for the solar physics and machine learning communities*
|
|
|
|
| 38 |
|
| 39 |
**FlareBench** is a novel benchmark dataset for solar flare prediction that covers the entire 11-year solar activity cycle. This dataset was introduced in our ICCV 2025 paper "Deep Space Weather Model: Long-Range Solar Flare Prediction from Multi-Wavelength Images".
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
### Why FlareBench?
|
| 42 |
|
| 43 |
Most conventional datasets for solar flare prediction do not cover diverse solar activity states. Consequently, models trained on such datasets can exhibit biases towards specific periods of solar activity. Furthermore, many datasets contain only small, low-resolution sunspot patches.
|
|
|
|
| 54 |
├── solar_images/
|
| 55 |
│ ├── aia/ # Atmospheric Imaging Assembly (9 wavelengths)
|
| 56 |
│ │ ├── 2011/ # 📊 Data available (Apr-Dec 2011)
|
| 57 |
+
│ │ ├── 2012/ # 📂 Same structure as 2011, no data files
|
| 58 |
│ │ ├── ...
|
| 59 |
│ │ └── 2024/
|
| 60 |
│ └── hmi/ # Helioseismic and Magnetic Imager
|
| 61 |
│ ├── 2011/ # 📊 Data available (Jan-Dec 2011)
|
| 62 |
+
│ ├── 2012/ # 📂 Same structure as 2011, no data files
|
| 63 |
│ └── ...
|
| 64 |
└── xrs/ # X-Ray Sensor data from GOES satellites
|
| 65 |
├── 2011/ # 📊 Data available (full year)
|
| 66 |
+
├── 2012/ # 📂 Same structure as 2011, no data files
|
| 67 |
└── ...
|
| 68 |
```
|
| 69 |
|
|
|
|
| 109 |
- Intended use of the dataset
|
| 110 |
- Preferred data transfer method
|
| 111 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
## 📚 Citation
|
| 113 |
|
| 114 |
If you use FlareBench in your research, please cite our ICCV 2025 paper:
|
|
|
|
| 146 |
- **Institution**: Keio University
|
| 147 |
- **Project Page**: [https://keio-smilab25.github.io/DeepSWM](https://keio-smilab25.github.io/DeepSWM)
|
| 148 |
|
|
|
|
|
|
|
|
|